Advanced Process Control Conference in Mesa, Arizona given by Alan Weber

1. www.cimetrix.com
28th Advanced Process Control Conference
Mesa, Arizona October 17-20, 2016
Alan Weber – Cimetrix Incorporated
The Role of Models in
Semiconductor Smart Manufacturing
1

2.  What is “Smart Manufacturing?”
 SEMI Standards evolution
 Equipment model examples
 Factory application use cases
 Conclusions
Outline
2

3.  “… cyber-physical systems monitor physical processes,
create a virtual copy of the physical world and make
decentralized decisions.
 Over the Internet of Things, cyber-physical systems
communicate and cooperate with each other and with
humans in real time…”
What is “Smart Manufacturing?”
From Industry 4.0 Wikipedia…
3

4. Imagine the collaborative behavior that could emerge !
 Discoverable
 Autonomous
 Model-based
 Secure
 Self-monitoring
 Easy to use
 Compact
 Standards-based
Components of a Smart Factory
Attributes of all these connected “things”
4

5.  Equipment models are useful
 Help understand equipment and process behavior
 Improve communication with suppliers
 Explicit, standard models are especially useful
 Support generic applications across equipment types
 Enable performance benchmarking within/among fabs
 Events and associated data offer untapped benefit
 Time lost can never be recovered
 Time is the ultimate unifying concept
 Models are the basis for component interoperability
 From basic sensors to intelligent subsystems
Key messages
5

6. Natural language analogy…
 SECS-I vocabulary (data items)
 SECS-II grammar (streams/functions)
 GEM sentences (capabilities)
 GEM300 conversations (scenarios)
 EDA improv theatre (dynamic DCPs)
 E164 improv theatre with a point (common model)
 <tbd> spontaneous flash mob (IIoT, …)
Evolution of equipment models
Referenced in SEMI standards
6

7. SEMI EDA equipment metadata model
Structure, nodes, self-description services
7

8. The equipment model value chain
EDA
Model
High-
Volume
Factory Ops
Pilot Factory
Operations
Process
Engineering
Equipment
Developers
Equipment
Components
Cimetrix
Software
Standard
Model
KPIs (metrics)
• Time to money
• Yield
• Productivity
• Throughput
• Cycle time
• Capacity
• Scrap rate
• EHS
Control Connect Collaborate Visualize Analyze Optimize
8

9. • Model structure exactly reflects tool
hardware organization
• Complete description of all potentially
useful information in the tool
• Always accurate, always available – no
additional documentation required
• Common point of reference among tool,
process, and factory stakeholders
• Source of unambiguous identifiers/tags
for database [auto] configuration
• Enables “plug and play” applications
Equipment metadata model benefits
Standards for all equipment models Process Module #1
Gate Valve Data
Substrate Location
Utilization
More Data,
Events, Alarms
Process Tracking
Other
Components
9

10.  What is “Smart Manufacturing?”
 SEMI Standards evolution
 Equipment model examples
 Factory application use cases
 Conclusions
Outline
10

11.  OEE calculation
 Substrate tracking
 Process execution tracking
 Lot completion estimation
 Product time measurement
 Fault detection/classification
 External sensor integration
 Component fingerprinting
 Others…
Example model-based applications
In general order of increasing complexity…
11

12. Substrate tracking
E90 state machines and model content
12

13. Process execution tracking
E157 state machine, model content, and results
13

14.  Motivation
 Inter-process wait times have direct negative impact on yield for
critical process steps
 Many advanced processes include a number of direct tool-to-tool
material delivery steps
 Productivity KPIs are also affected by inaccurate carrier completion
estimates
 Requirements
 Provide continuously updated estimates for current lot completion
and equipment idle time for MCS/AMHS dispatching decision support
 Provide notification events at configurable thresholds
 Maintain substrate process times per recipe
Lot completion estimation
Motivation and requirements
14

15.  Sum # of wafers to be processed
 For each Carrier SEMI Object instance select ControlJobs with
CarrierInputSpec that contains Carrier’s ObjID
 For each ControlJob, count the # of substrates listed in each ProcessJob’s
PRMtlNameList attribute
 Calculate average time to return substrate to destination carrier
 Record time when first AtWork-AtDestination event is reached
 When next AtWork-AtDestination event is reached, record difference as
current average time to return substrate to carrier
 Calculate initial carrier completion estimation
 = # remaining substrates * current average substrate return time
 Update carrier completion estimation
 When each AtWork-AtDestination event is reached, subtract timestamp of
first event from latest event, and divide by # of substrates
 Use this value as new average substrate return time in calculating new
carrier completion estimation
Lot completion estimation
Algorithm summary
15

16. Equipment model content
Used in lot completion estimation algorithm (1)
Carrier ObjID
attribute
High-level
Equipment
structure
E90 Substrate
Transport
events
MaterialManager
Module
16

17. Equipment model content
Used in lot completion estimation algorithm (2)
High-level
Equipment
structure
JobManager
Module
ControlJob
CarrierInputSpec
attribute
ProcessJob
PRMtlNameList
attribute
17

18.  Description
 Capture locations visited by each substrate during lot
processing, with absolute and relative timestamps
and durations for each
 Use knowledge of process module execution status to
calculate “active” and “wait” time elements for each
substrate (now standardized in SEMI E168)
 KPIs affected
 Equipment productivity by identifying unnecessary
wait times and understanding/addressing the
associated root causes
 Process variability through increased tool behavior
visibility
 EDA model leverage
 Substrate tracking events directly support this
function but are not usually collected sufficiently
using GEM to support this need
 All other events required to classify all time segments
in a substrate’s life cycle are mandated by metadata
model standards
Model-based application examples
Product Time Measurement (Wait time waste analysis)
18

19. How much data is required?
The more data, the greater the benefit
Data Used Analysis Enabled Benefit / ROI
Lot start/stop events
(from MES)
Average throughput calculations by product and
process
Establish performance baseline
Lot start/stop events
(from equipment)
Actual throughput rates by product, recipe, and
tool
Identify bottlenecks and laggards among toolset
Chamber-level wafer process
start/stop events
Actual throughput rates by product, recipe, and
chamber; tool-level wait time per wafer
Identify areas for improvement in equipment and
process engineering
Wafer transport and location status
events within equipment
Details of wait and active time per wafer Identify recipe and equipment design issues
Other component-level signals
related to wafer movement
Fine-grained analysis of wafer movement
behavior
Identify even more recipe and equipment
design/calibration issues
AMHS carrier movement and
storage events
Wait and active (transport) time between tools;
door-to-door value chain analysis
Identify areas for improvements needed in production
scheduling, dispatching, and operations
19

20. Data Collection Level vs. Benefit
Revenue improvement potential
0 1 2 3 4 5
Lot-level
tool data
Wafer-level
chamber
data
Wafer-level
component
data
Carrier-level
AMHS data
Data Collection Level
Delta
Revenue %
20

21.  Description
 Multivariate statistics used to develop reduced-dimension
equipment fault models for various operating points
 Fault models evaluated in real-time to detect process drift
and/or impending tool failure
 May interdict tool operation in mid-run to prevent/reduce
scrap
 KPIs affected
 Process yield and scrap rate through higher detection
sensitivity
 Equipment availability resulting from fewer false positives
 EDA model leverage
 Conditional triggers in trace request frame data collection
 Metadata model contains context to select proper fault
detection algorithms
 Multi-client access to develop, evaluate, and update fault
models
Model-based application examples
Multivariate Fault Detection and Classification (FDC)
21

22.  Description
 Create “sockets” in equipment model for common external
sensors, key subsystems, and associated context information
 Address most of the key challenges in external sensor
integration using shared metadata model and EDA services to
associate context and timing information close to the data
source
 Directly applicable to integration of sub-fab components
(pumps, chillers, scrubbers, abatement systems)
 KPIs affected
 Process capability and yield through improved control capability
 Engineering efficiency through reduced custom integration
effort
 EDA model leverage
 Shared metadata model aggregates data from multiple, diverse
data sources, presenting unified perspective to client
applications
 Multi-client capability enables experimentation with additional
sensors and control techniques without disrupting production
Model-based application examples
External sensor/subsystem integration Sensor Integration Challenges
1. Finding a sensor that works
2. Sampling/process synchronization
3. Dealing with multiple timestamps
4. Scaling and units conversion
5. Applying factory naming convention
6. Associating context and sensor data
7. Ensuring statistical validity
8. Aligning results in process database
22

23. External sensor integration example
With challenge areas addressed
OEM Tool
EDAGEM
Sensor/Actuator Gateway
(pumps, chillers, scrubbers)
Device Drivers
DP ChTP Sc
TPPump
I/F
FICS / MES
EDA Client
EDA Server
EDA Client
Smar
t
Data
Model
Raw Data
Metadata Model
Public
Data
DCIM*
DCI
M
Proprietary
Application
s
Process-specific
applications Factory-level
EDA Client Apps
(DOE, FDC, PHM, …)
Custom
or
EtherCAT
TCP/IP
HTTP
HTTP HTTP
To factory-level systems
Context data
Synchronization data
S1 S2
* DCIM =
Data Collection
Interface Module
Synchronization signals
Process
Engineering
Database
2
3
4
5
6
7
1
8
23

24.  Models have been at the core of SEMI's
Information and Control standards for decades
 Models help components of a smart manufacturing
system understand one another
 The sophistication of these models is now sufficient
to support true application interoperability
 Industry standard models can greatly reduce
factory application development cost
 Models will play an increasingly important role as
the number and variety of components multiply
Conclusions
24